Structural basis for activation of DNMT1

DNMT1 is an essential enzyme that maintains genomic DNA methylation, and its function is regulated by mechanisms that are not yet fully understood. Here, we report the cryo-EM structure of human DNMT1 bound to its two natural activators: hemimethylated DNA and ubiquitinated histone H3. We find that a hitherto unstudied linker, between the RFTS and CXXC domains, plays a key role for activation. It contains a conserved α-helix which engages a crucial “Toggle” pocket, displacing a previously described inhibitory linker, and allowing the DNA Recognition Helix to spring into the active conformation. This is accompanied by large-scale reorganization of the inhibitory RFTS and CXXC domains, allowing the enzyme to gain full activity. Our results therefore provide a mechanistic basis for the activation of DNMT1, with consequences for basic research and drug design.

DNA methylation is a key epigenetic mark that regulates gene expression and genome stability 1,2 . In mammals, DNA methylation occurs at the 5 th position of cytosine, mostly within CpG dinucleotides, and it is catalyzed by a DNA methyltransferase (DNMT) family. De novo DNMTs (DNMT3A, 3B, and 3L) set up proper DNA methylation pattern during development and differentiation, and this pattern is faithfully copied on the newly replicated DNA at each round of cell division, by the maintenance enzyme DNMT1 3,4 . An E3 ubiquitin ligase (ubiquitinlike containing PHD and RING finger domains 1, UHRF1) protein, plays a crucial role for maintenance DNA methylation 5,6 , together with DNMT1. UHRF1 recognizes hemimethylated DNA via its SET and RINGassociated (SRA) [7][8][9] and catalyzes double monoubiquitination at K18 and K23 on histone H3 (H3Ub2) which, in turn, recruits DNMT1 and stimulates its enzymatic activity [10][11][12][13][14] .
DNMT1 is a large protein (1616 amino acids), containing multiple domains (Fig. 1a), and subject to intramolecular regulations that strongly restrict its activity to hemimethylated DNA 15 . In the absence of DNA (apo-DNMT1, PDB:4WXX, aa:351-1600), the enzyme is autoinhibited: Binding of Replication-Foci Targeting Sequence (RFTS) to the catalytic core, in association with recognition of the DNA binding region by an Auto-Inhibitory Linker, inhibit the access of hemimethylated DNA to DNMT1 catalytic region 16,17 . A key unresolved question is: how does the combined presence of H3Ub2 and hemimethylated DNA allow the enzyme to overcome this double inhibition? Of note, previous structural studies of DNMT1 in a complex with hemimethylated DNA (PDB:4DA4, aa:731-1602; 6X9I, aa:729-1600) [18][19][20] , in a complex with unmethylated CpG DNA (PDB:3PTA, aa:646-1600) 21 and RFTS bound to H3Ub2 (PDB:5WVO, aa:351-600; 6PZV, aa:349-594) 12, 13 have used a truncated version of the protein (Fig. 1a), therefore the fate of the inhibitory regions, RFTS and Auto-Inhibitory Linker, during activation is unknown. In order to understand the detailed molecular mechanism for DNMT1 activation, we have determined the cryogenic electron microscopy (cryo-EM) structure of human DNMT1 (aa:351-1616), stimulated by the H3Ub2 tail and in an intermediate complex with a hemimethylated DNA analog. Our structure illuminates the synergistic structural rearrangements that underpin the activation of DNMT1. In particular, it highlights the key role of a hydrophobic "Toggle" pocket in the catalytic domain, which stabilizes both the inactive (inhibited) or the active states. In the latter case, it functions by accepting a pair of phenylalanine residues from a hitherto unrecognized, yet highly conserved Activating Helix located between the RFTS and CXXC domains.

Results
Cryo-EM structure of DNMT1 bound to ubiquitinated H3 and hemimethylated DNA To uncover the molecular mechanism of DNMT1 activation, we conducted cryo-EM single particle analysis of DNMT1 (aa:351-1616) in an intermediate complex with H3Ub2 tail and hemimethylated DNA (Fig. 1a, b). The human DNMT1 protein, a minimum fragment required for investigating the activation mechanism by binding of ubiquitinated H3, was produced using the Sf9 baculovirus expression system, and an H3 tail peptide (aa:1-37 W, K14R/K27R/K36R) was dimonoubiquitinated on K18 and K23 to completion in vitro (Supplementary Fig. 1a and see method). As expected from previous work 12, 13 , in contrast to K18 or K23 single monoubiquitinated H3, the addition of H3Ub2 effectively enhanced the enzymatic activity of DNMT1 ( Supplementary Fig. 1d, e). The DNMT1:H3Ub2 binary complex was used for reaction with hemimethylated DNA. The target cytosine in hemimethylated DNA was replaced by a 5-fluorocytosine (5fC) to form an irreversible covalent complex with DNMT1 22 . The ternary complex containing DNMT1 bound to H3Ub2 and DNA mCG/fCG was purified by gel-filtration chromatography ( Supplementary Fig. 2a-c), and used for cryo-EM single particle analysis ( Supplementary Fig. 3a).
3D variability analysis by cryoSPARC 23 revealed 2 types of particles: Type I particles with 2.5 Å resolution, in which the CXXC domain was ordered ( Fig. 1b and Supplementary Fig. 4), and Type II particles with 2.2 Å resolution, in which the CXXC was disordered (Supplementary Fig. 5a). The atomic models of DNMT1 were constructed from the two types (Supplementary Table 1); these structures were essentially the same (except for the CXXC domain), therefore in the rest of the paper we will focus on the Type I DNMT1 particle, with the ordered CXXC. In this structure, the RFTS domain, Auto-Inhibitory Linker, N-terminal β-sheet of BAH1 (aa:731-755), and some loops and linkers were invisible, reflecting their flexibility (Fig. 1b). The structured elements of the ternary complex showed that the catalytic core of DNMT1 bound to hemimethylated DNA, the catalytic loop (aa:1224-1238) recognized flipped-out 5fC, and the Target Recognition Domain (TRD) residues (Cys1499, Leu1500, Trp1510, Leu1513, Met1533, and Gly1534) bound to the methyl-group of 5mC ( Supplementary Fig. 5b-d). Overall, these results confirm previously reported DNMT1:hemimethylated DNA binary complex structures (Supplementary Notes) [18][19][20] . In addition, they show the behavior of the RFTS and Auto-Inhibitory Linker in the active form, as described in the following section.

Spatial rearrangement of RFTS and CXXC domains in the active form
The dissociation of the RFTS domain from the catalytic core is assumed to be required for DNMT1 activation, yet no direct evidence of this structural rearrangement has been shown to date. Our cryo-EM structure of apo-DNMT1 (aa:351-1616) at 3.4 Å resolution showed a fully-structured RFTS domain bound to the catalytic core (Supplementary Fig. 6a left, and Supplementary Table 2). In contrast, the cryo-EM map of the DNMT1:H3Ub2:DNA mCG/fCG ternary complex showed no density for the RFTS:H3Ub2 (Fig. 1b). Intriguingly, particles smaller than the DNMT1:H3Ub2:DNA mCG/fCG ternary complex were observed in 2D class average ( Supplementary Fig. 7a). The size and shape of these particles were comparable to those of RFTS:H3Ub2 as estimated from the 2D projected template of RFTS:H3Ub2 (PDB:5WVO)-derived 3D Gaussian model ( Supplementary Fig. 7b). These data indicate that the RFTS:H3Ub2 moiety is in a highly dynamic state and does not interact with other domains when DNMT1 is active.
To separate the contributions of H3Ub2 and of hemimethylated DNA for displacement of the RFTS, we determined the cryo-EM structure of DNMT1 (aa:351-1616) in complex with H3Ub2, but without DNA (Supplementary Figs. 1b, 6a right, and Supplementary Table 2). This structure reached 3.6 Å resolution and showed that, while the N-lobe of DNMT1 became flexible and therefore invisible, the C-lobe remained bound firmly to the catalytic core upon H3Ub2 binding. These data were further supported by small angle X-ray scattering (SAXS) analyses (Supplementary Figs. 6b, 8a-c, Supplementary Table 3, and Supplementary Notes), showing that H3Ub2, in itself, is not sufficient to dislodge the RFTS.
We next investigated the dynamics of the CXXC domain after DNMT1 activation. In apo-DNMT1, the CXXC domain is affixed to the side of the RFTS domain (Fig. 2a left) 16,17 . In the presence of unmethylated DNA (DNA CG/CG ) (PDB:3PTA), the zinc finger motif of the CXXC domain recognizes unmethylated CpG, shifts 30 Å towards the TRD domain, and sits near the active center (Fig. 2a right) 21 . In this conformation, the Auto-Inhibitory Linker directly interrupts the binding of DNA to the active site, which prevents unlicensed de novo DNA methylation. Intriguingly, in our active complex, the CXXC domain moved away from the active center and took an "upside-down" position (relative to 3PTA), between the BAH1 domain and the catalytic domain with which it established hydrogen bonds and van der Waals interactions (Fig. 2b, c).
As the DNA binding interface of the CXXC remained solventexposed in the active complex ( Fig. 2a center), so we asked whether it could still bind unmethylated DNA. However, the superimposition of the CXXC:DNA complex onto the ternary complex structure revealed major steric clashes with the catalytic domain (Fig. 2d), suggesting that DNA binding by the CXXC is fully suppressed when DNMT1 is active. The new position of the CXXC leads to a structural deformation of the BAH1 N-terminal β-sheet and induces a different orientation of the BAH1 loop (aa:765-775) (Figs. 1b, 2a center). These structural changes then lead to full eviction of the Auto-Inhibitory Linker from the catalytic core and causes the Auto-Inhibitory Linker to adopt a highly disordered structure (Figs. 1b, 2a).
Taken together, our cryo-EM analysis of the active form of DNMT1 revealed a large reorganization of the inhibitory domains relative to unliganded or inactive forms. Upon joint binding of H3Ub2 and hemimethylated DNA, the RFTS domain is forced out of the catalytic core, while the CXXC domain undergoes a drastic spatial rearrangement, ultimately leading to the eviction of the Auto-Inhibitory Linker from the catalytic core.
A "Toggle" pocket accepts different phenylalanines in the repressed and active states Zooming in on the catalytic site, we observed another striking difference between the inactive and active states of the enzyme.
This contrasts sharply with our activated form of DNMT1 (Fig. 3c). In that situation, the Activating Helix is shortened as it forms a helixturn; its residues Phe631 and Phe632 invade the Toggle Pocket. The DNA Recognition Helix is freed from the Toggle Pocket and springs into a straight conformation, which allows (i) access of Phe1235/ Arg1238 (in the catalytic loop) to the minor groove at the mCG/fCG site, (ii) formation of a hydrogen bond between Tyr1240 in the DNA Recognition Helix and the phosphate backbone of the DNA, and (iii) engagement of Phe1243 (in the DNA Recognition Helix) by Pro615, Lys617, and Gln635 (in the linker between RFTS and CXXC domains), which prevents Phe1243 from entering the Toggle Pocket (Fig. 3a, c and Supplementary Fig. 9a).
Interestingly, a previous study of human DNMT1 binary complex (aa:729-1600, not containing Activating Helix) with hemimethylated DNA analog (active state, PDB: 6X9I) shows that Phe1243 remains in the Toggle Pocket and the N-terminal region of the DNA Recognition Helix structure is partially unfolded, thereby preventing the binding of Tyr1240 to the phosphate backbone of DNA ( Supplementary  Fig. 9a, b) 20 . Furthermore, folded or unfolded DNA Recognition Helix structures are observed in the previous structures of mouse DNMT1 (aa:731-1602) bound to hemimethylated DNA analog, depending on the sequence around the mCG/fCG 18,19 , suggesting that the DNA Recognition Helix is intrinsically flexible; in contrast, our DNMT1:H3Ub2:DNA mCG/fCG complex shows a rigid conformation of the DNA Recognition Helix, indicating that the phenylalanine pair in the Activating Helix contributes to the activation state of DNMT1.

Crucial role of a phenylalanine pair for activation of DNMT1
The phenylalanines Phe631 and Phe632 are invariant between vertebrate species, and are also present in the cephalochordate Amphioxus (Fig. 3d). We, therefore, asked whether these residues played a role in the activation of DNMT1, as could be expected from the fact that they bind the Toggle Pocket. In a binding assay, we found that the F631A/ F632A mutations abolished the ability of DNMT1 to bind hemimethylated DNA (Fig. 4a), even though the H3Ub2-binding ability of DNMT1 was unaffected ( Supplementary Fig. 1f). We carried out an in vitro DNA methylation assay and, again, observed that the F631A/ F632A mutation led to severe defects in DNA methylation (Fig. 4b).
We then sought confirmation of these results in an in vitro assay, which reconstitutes replication-coupled maintenance DNA methylation using Xenopus egg extracts 11,12,24 . In that system, we immunodepleted DNMT1, and re-introduced recombinant DNMT1, either WT or mutated on the two phenylalanines of the Activating Helix (F506A and F507A in Xenopus, FF/AA mutant). As previously reported, the depletion of xDNMT1 from Xenopus egg extracts resulted in the accumulation of chromatin-bound UHRF1 and ubiquitinated histone H3 species (Fig. 4c); this is due to defective maintenance DNA methylation, which generates hemimethylated DNA from which UHRF1 cannot be released 11,24 . The addition of wild-type (WT) recombinant xDNMT1 suppressed the accumulation of UHRF1 and ubiquitinated H3 (Fig. 4c). The FF/AA mutant retained chromatin binding activity but failed to suppress the accumulation of UHRF1 and ubiquitinated H3, showing defects in maintenance DNA methylation (Fig. 4c). Therefore, this functional assay in Xenopus egg extracts further validated the effect of the mutation.
Lastly, we used a colon cancer cell line HCT116, in which DNA methylation has been widely studied. In this line, both endogenous alleles of DNMT1 are tagged with an auxin-inducible degron (AID) (Fig. 4d). In this DNMT1-AID line, we introduced rescue vectors: one encoding WT DNMT1, and the other encoding the FF/AA mutant (Fig. 4d). The level of endogenous DNMT1, exogenous WT or FF/AA DNMT1 were comparable (Fig. 4e), and they were located in the nucleus ( Supplementary Fig. 10). Treating the cells with indole-3-acetic acid (IAA) caused the disappearance of endogenous DNMT1 but did not affect the exogenous proteins. We then measured global DNA methylation at days 0, 4, and 8 after endogenous DNMT1 removal (Fig. 4f). The control cells (empty vector) lost almost one-third of total DNA methylation, going from 70 to 50% methylation. This loss was completely prevented by the WT DNMT1 transgene. In contrast, the FF/ AA DNMT1 mutant was incapable of sustaining DNA methylation maintenance and showed DNA methylation values close to those of the empty vector (Fig. 4f).
Collectively these experiments confirm that the Activating Helix, and especially its conserved phenylalanines, are crucial for DNA methylation maintenance by DNMT1.

Discussion
Our cryo-EM analysis reveals a molecular mechanism for human DNMT1 catalytic activation (Fig. 5). Our results reveal both the large-scale displacements of inhibitory modules (RFTS, CXXC, Auto-Inhibitory Linker), as well as more detailed changes, particularly the switch by which the same hydrophobic pocket, initially bound to inhibitory phenylalanines, engages activating phenylalanines, which releases the DNA Recognition Helix and permits catalysis. This regulation also operates in Xenopus, and may even occur in invertebrates such as Amphioxus, in which the regulatory amino acids are conserved (Fig. 3d).
The catalytic domains of the de novo DNA methyltransferases DNMT3A and DNMT3B bind the DNMT3L catalytic-like domain and form a heterotetramer [25][26][27][28] . Interestingly, the DNMT3A(B)/3 L interface is formed by hydrophobic interactions mediated by phenylalanine residues, and therefore is known as the F-F interface. The F-F interface enhances DNA methylation activity by the DNMT3A(B)/3L heterotetramer 25 . The hydrophobic residues in the DNMT3A(B) catalytic domain spatially corresponds to the Toggle Pocket of DNMT1. Thus, covering the hydrophobic pocket of the catalytic domain by an intra-or inter-molecule interaction could be an evolutionarily conserved activation mechanism of DNA methyltransferases. The Activating Helix, however, is unique to DNMT1 and crucial for enzymatic activation, and therefore could be utilized to design novel inhibitors such as helical peptides that mimic this Activating Helix.
Previously reported structures of apo-DNMT1 and DNMT1:DNA CG/CG revealed a dual-auto-inhibitory mechanism in which the RFTS domain and the Auto-Inhibitory Linker are embedded into the catalytic core, thereby inhibiting the access of cognate DNA (Fig. 5). Our cryo-EM analysis of the ternary complex showed full dissociation of both the RFTS domain and Auto-Inhibitory Linker from the catalytic core (Fig. 1b). Interestingly, H3Ub2-binding to the RFTS domain might not be sufficient for the displacement of the RFTS domain as our cryo-EM and SAXS data showed that the C-lobe of RFTS domain is still accommodated in the catalytic core in the DNMT1:H3Ub2 complex (Supplementary Fig. 6). A previous molecular dynamics simulation has demonstrated that H3Ub2-binding reduces the contact number between the C-lobe and catalytic core 12 . In addition, apo-DNMT1 was unable to form the binary complex with hemimethylated DNA (Supplementary Fig. 1c). We hypothesize, therefore, that H3Ub2 binding destabilizes the inhibitory interaction between the C-lobe and the catalytic core, allowing hemimethylated DNA to penetrate the catalytic core. Thus, we propose that simultaneous binding to H3Ub2 and DNA leads to full activation of DNMT1 via the following structural changes: (i) dissociation of the RFTS domain from the catalytic core, (ii) structural changes to an Activating Helix causing the conserved residues Phe631 and Phe632 to invade the Toggle Pocket of the catalytic domain, (iii) adoption of a rigid conformation by the DNA Recognition Helix, (iv) spatial rearrangement of the CXXC domain, and (v) eviction of the Auto-Inhibitory Linker from the catalytic domain (Fig. 5). However, it is currently unknown how simultaneous binding of H3Ub2 and hemimethylated DNA causes a conformational change in the Activating Helix to place Phe631 and Phe632 in the Toggle Pocket. Future work, such as molecular dynamics simulation, will determine if these structural changes occur sequentially or simultaneously. Thus, our findings describe concepts and mechanisms in the multi-step activation process of DNMT1 that ensures faithful maintenance of DNA methylation.

Protein expression and purification
The gene encoding wild type and mutant of human DNMT1 (residues 351-1616) containing N-terminal ten histidine tag (His-tag) and human rhinovirus 3 C (HRV 3 C) protease site was amplified by PCR and cloned into the pFastBac vector (Invitrogen) using the seamless cloning

Cryo-EM data collection
A 3 µL of the protein solutions was applied onto the glow-discharged holey carbon grids (Quantifoil Cu R1.2/1.3, 300 mesh). The grids were plunge-frozen in liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific). Parameters for plunge-freezing were set as follows: blotting time, 3 s; waiting time, 0 sec; blotting force, 0; humidity, 100%; and chamber temperature, 4°C. Data for DNMT1:H3Ub2:DNA mCG/fCG ternary complex was collected at the University of Tokyo on a 300 kV Titan Krios electron microscope (Thermo Fisher Scientific) with a K3 direct electron detector (Gatan) with BioQuantum energy filter in counting mode. A total of 4068 movies were recorded at a nominal magnification of ×105,000 and a pixel size of 0.83 Å/pixel, with a defocus range between −0.8 and −1.8 μm and a dose rate of 1.25 electrons/Å 2 per frame. The data were automatically acquired using the SerialEM 3.9.0 software 29 . A typical motion-corrected cryo-EM image is shown in Supplementary Fig. 11a.
Data of apo-DNMT1, DNMT1:H3Ub2 were collected at RIKEN BDR on a 200-kV Tecnai Arctica electron microscope (Thermo Fisher Scientific) with a K2 direct electron detector (Gatan) in counting mode. A total of 2071 movies for apo-DNMT1 and 1869 movies for DNMT1:H3Ub2 were recorded at a nominal magnification of ×23,500 and a pixel size of 1.477 Å/pixel, with a defocus range between −0.8 and −1.4 μm, and a dose rate of 1.25 electrons/Å 2 per frame. A typical motion-corrected cryo-EM image is shown in Supplementary Figs. 11b, c. The data were automatically acquired using the SerialEM 3.8 software.

Data processing
All data were processed using cryoSPARC v3.2.0 30 for PDB deposition. The movie stacks were motion corrected by Full-frame motion correction or Patch motion correction. The defocus values were estimated from Contrast transfer function (CTF) by Patch CTF estimation or CTFFIND4 31 . A total of 4,307,107 particles of DNMT1:H3Ub2:DNA mCG/fCG ternary complex were automatically picked using a blob picker with 80, 105, and 130 Å circular blobs and 80-130 Å elliptical blobs (Supplementary Fig. 11a). Particles (3,798,046) were then extracted in a box size of 256 pixels with a 0.83 Å/pixel size followed by a single round of reference-free 2D classifications (Supplementary Fig. 11a). The selected good class containing 1,621,988 particles were used for ab initio 3D reconstruction (Supplementary Fig. 3a). Then, non-uniform refinement was performed against all the extracted particles to yield the cryo-EM map with an overall resolution of 2.09 Å resolution. The subsequent heterogeneous refinement selected 2,653,627 particles as a good class. These particles were subjected to a 3D variability analysis, separating the CXXC-ordered and CXXC-disordered models. The particles (138,662) in the CXXC-ordered model and those (897,446) in the CXXC-disordered models were then subjected to non-uniform refinement to generate a cryo-EM map with an overall resolution of 2.52 and 2.23 Å, respectively. The classification processes were shown in Supplementary Fig. 3a, and the statics of data collection and refinement, and validation were shown in Supplementary Table 1. A total of 3,984,637 particles of apo-DNMT1 were automatically picked using a blob picker with 80, 105, and 130 Å circular blobs ( Supplementary Fig. 11b). Particles (3,984,637) were then extracted in a box size of 160 pixels with a 1.477 Å/pixel size, followed by two rounds of reference-free 2D classifications to remove junk particles (Supplementary Fig. 11b). The selected 3,666,067 particles were subjected for ab initio 3D model reconstruction to generate four cryo-EM maps ( Supplementary Fig. 3c). An initial model shows similar shape with the crystal structure of DNMT1 (PDB: 4WXX). Then, non-uniform refinement was performed against the particles classified in the initial model of DNMT1 (1,824,727). These particles were re-extracted in a box size of 256 pixels with a 1.477 Å/pixel size by local motion correction. Nonuniform refinement yields the cryo-EM map with an overall resolution of 3.32 Å resolution. These particles were subjected to a 3D variability analysis and heterogeneous refinement, removing the dimer particles, and separating the RFTS-free map and RFTS-bound map. To improve the cryo-EM map, further 3D variability analysis and clustering by PCA was performed. The particles (380,989) were then subjected to nonuniform refinement to yield the final cryo-EM map. The overall resolution of 3.45 Å resolution using the gold-standard Fourier shell correlation with a 0.143 cut-off. The classification processes were shown in Supplementary Fig. 3c, and the statics of data collection and refinement and validation were shown in Supplementary Table 2. A total of 2,463,410 particles of DNMT1:H3Ub2 were automatically picked using a blob picker with 80, 105, and 130 Å circular blobs (Supplementary Fig. 11c). The particles were then extracted in a box size of 160 pixels in a 1.477 Å/pixel size using cryoSPARC followed by initial dataset cleanup using reference-free 2D classifications ( Supplementary  Fig. 11c). The selected 2,336,267 particles were subjected for ab initio 3D model reconstruction to generate four cryo-EM maps ( Supplementary  Fig. 3b). Further 2D classifications were performed by the 2 class of ab initio 3D model assigned as DNMT1 particles. The particles (160,088) belonging to the best four 2D classes were subjected to create the fine initial model. For further refinement, the particles (1,303,645) without ice images from 2D classification II were selected. These particles were re-extracted in a box size of 256 pixels with a 1.477 Å/pixel size by local motion correction. Non-uniform refinement yields the cryo-EM map with an overall resolution of 3.55 Å resolution. These particles were subjected to a 3D variability analysis and heterogeneous refinement, separating the RFTS-free map and the RFTS-bound map. The subsequent 2D classification selected 735,233 particles as RFTS-bound structures. To improve the cryo-EM map, further 3D variability analysis and clustering by PCA was performed. The particles (645,368) were then subjected to non-uniform refinement to yield the final cryo-EM map. The overall resolution of 3.52 Å resolution using the gold-standard Fourier shell correlation with a 0.143 cut-off. The classification processes were shown in Supplementary Fig. 3b, and the statics of data collection and refinement and validation were shown in Supplementary Table 2.
For the analysis of the H3Ub2-RFTS domain complex, the 2D classification analysis were also performed by RELION-3.1 (Supplementary Fig. 7) 32 . The movie stacks of DNMT1:H3Ub2:DNA mCG/fCG were motion corrected by MotionCor2. The defocus values were estimated from CTF by CTFFIND4 31 . Single particle image was also extracted by LoG Auto picker of RELION to check the particles of other biomolecules. After three rounds of 2D classification, the particle smaller than DNMT1 was selected. These smaller particles (80,186) were reextracted in a box size of 128 pixels with a 0.83 Å/pixel size, and the images were classified by 2D classification. The major 2D average images were compared with the projected templates of the RFTS-H3Ub2 complex (PDB: 5WVO) (Supplementary Fig. 7). Gaussian model of the complex was created by the Molmap of ChimeraX 33 . The 2D projected templates were created by the module "create template" in cryoSPARC.

SEC-SAXS
SAXS data were collected on Photon Factory BL-10C using an HPLC Nexera/Prominence-I (Shimazu) integrated SAXS set-up. About 100 µL of 10 mg/mL of the apo-DNMT1 (aa:351-1616) or its bound to H3Ub2 S-S were loaded onto a Superdex® 200 Increase 10/300 GL (Cytiva) preequilibrated with 20 mM Tris-HCl (pH 8.0), 150 mM NaCl and 5% glycerol at a flow rate of 0.5 mL/min at 20°C. The flow rate was reduced to 0.05 mL/min at an elution volume of 10−13 mL. X-ray scattering was collected every 20 s on a PILATUS3 2 M detector over an angular range of q min = 0.00690 Å −1 to q max = 0.27815 Å −1 . UV spectra at a range of 200 to 450 nm were recorded every 10 s. Circular averaging and buffer subtraction were carried out using the program SAngler 34 to obtain one-dimensional scattering data I(q) as a function of q (q = 4πsinθ/λ, where 2θ is the scattering angle and λ is the X-ray wavelength 1.5 Å). The scattering intensity was normalized on an absolute scale using the scattering intensity of water 35 . The multiple concentrations of the scattering data around the peak at A280, namely ascending and descending parts of the chromatography peak, and I(0) were extrapolated to zero-concentration by Serial Analyzer 36 . The molecular mass of the measured proteins was estimated by the empirical volume of correlation, V c , showing no aggregation of the measured sample 37 . The radius of gyration R g and the forward scattering intensity I(0) were estimated from the Guinier plot of I(q) in the smaller angle region of qR g < 1.3. The distance distribution function P(r) was calculated using the program GNOM 38 . The maximum particle dimension D max was estimated from the P(r) function as the distance r for which P(r) = 0. The scattering profile of the crystal structure of apo-DNMT1 and its docking model with ubiquitinated H3 were computed with CRYSOL 39 .

In vitro DNA methylation assay
The 42 base pair of DNA duplex containing three hemimethylation sites (0−0.8 µM) was methylated with the recombinant DNMT1 (15 nM, aa:351-1616) by the addition of the disulfide-linked ubiquitinated H3 (1 µM H3Ub2 S-S ) including 20 µM SAM in reaction buffer (20 mM Tris-HCl [pH 8.0], 50 mM NaCl, 1 mM EDTA, 3 mM MgCl 2 , 0.1 mg/mL BSA, and 20% Glycerol) at 37°C for 1 hr. Termination of methylation reaction and conversion of SAH to ADP were performed by the addition of 5×MTase-Glo TM reagent from methyltransferase assay kit, MTase-Glo (Promega) at 1:4 ratio for the reaction total volume. After 30 min stationary at room temperature, the ADP detection process was carried out with solid white flat-bottom 96-well plates (Costar). MTase-Glo TM Detection Solution was added to the reaction in a 1:1 ratio to reaction total 40 µL volume and incubated for 30 min at room temperature. The luminescence derived from the reaction product, SAH, was monitored using GloMax® Navigator Microplate Luminometer (Promega). The effect of F631A/F632A mutation were examined at the condition of the DNMT1 (15 nM) with 1 µM H3Ub2 S-S by the addition of 42 base pair of DNA duplex (0−0.8 µM) in the same reaction buffer. The SAH conversion process and ADP detection process are in the manner described above.
For the evaluation of the DNMT1:H3Ub2 complex, the final concentration of DNMT1 was 50 nM to prevent the dissociation of ubiquitinated H3 (K D = 18 nM). DNA methylation reactions were initiated by mixing of apo-DNMT1 or DNMT1:H3Ub2 iso and stopped at 0, 5, 15, or 30 min by addition of 5×MTase-Glo TM reagent. The detection process was performed in the same way as described above. At least three independent experiments were performed for the estimation of standard deviation.

DNA pull-down assay
About 20 µg of the 21-base pair of biotinylated hemimethylated DNA duplex was immobilized on Dynabeads M-280 Streptavidin (VERITAS) equilibrated with the binding buffer (20 mM Tris-HCl [pH 7.5], 150 mM NaCl, 10% Glycerol, and 0.05% Nonidet P-40 (NP-40)). After washing the beads with the binding buffer, 10 µg of purified DNMT1 (aa:351-1616) wild-type or F631A/F632A mutant, 2-equimolar excess of H3Ub2 S-S and equimolar of SAH were added to the beads. After incubation for 2 hrs at 4°C, the unbound proteins were washed five times with the binding buffer. The proteins bound to the immobilized DNA were boiled for 2 min at 95°C in an oxidative SDS-loading buffer and analyzed by SDS-PAGE using SuperSep TM Ace, 5-20% gel (Wako, Japan). At least three independent experiments were performed.

Cell culture, transfection, and colony isolation
The HCT116 cell line, which conditionally expressed OsTIR1 under the control of a tetracycline (Tet)-inducible promoter, was obtained from the RIKEN BRC Cell Bank (http://cell.brc.riken.jp/en/), and genotyped by Eurofins. All cell lines were cultured in McCoy's 5 A medium (Sigma-Aldrich) supplemented with 10% FBS (Gibco), 2 mM L-glutamine, 100 U/mL penicillin, and 100 µg/mL streptomycin. Cells were grown in a 37°C humid incubator with 5% CO 2 . To generate stable DNMT1-AID cell lines, we followed previous studies 41,42 . Briefly, cells were grown in a 24-well plate, then CRISPR/Cas and donor plasmids were transfected using Lipofectamine 2000 (Thermo Fisher Scientific). Two days after transfection, cells were transferred and diluted in 10 cm dishes, followed by selection in the presence of 700 mg/mL G418 or 100 mg/mL Hygromycin B. After 10-12 days, colonies were picked for further selection in a 96-well plate. To induce the degradation of AID-fused proteins, cells were incubated with 0.2 µg/mL doxycycline (Dox) and 20 µM auxinole for 1 day, then we replaced the medium including 0.2 µg/mL Dox and 500 µM indole-3-acetic acid (IAA), a natural auxin.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
The data that support this study are available from the corresponding author upon reasonable request. The cryo-EM density map has been deposited in the Electron Microscopy Data Bank (EMDB, www.ebi.ac. uk/pdbe/emdb/) under accession code EMD-33200, EMD-33201, EMD-33298, EMD-33299, and the atomic coordinates of CXXC-ordered and CXXC-disordered ternary complex have been deposited in the PDB (www.rcsb.org) under accession code 7XI9 and 7XIB, respectively. All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. PDB 4WXX, 3PTA, 6X9I, 4DA4, and 5WVO were used for this study. Source data are provided with this paper.